MARINE ECOLOGY PROGRESS SERIES
Mar Ecol Prog Ser
Vol. 517: 229–250, 2014
doi: 10.3354/meps11043
Published December 15
FREE
ACCESS
Characterizing Pacific halibut movement
and habitat in a Marine Protected Area using net
squared displacement analysis methods
Julie K. Nielsen1,*, Philip N. Hooge2, 4, S. James Taggart2, 5, Andrew C. Seitz3
1
School of Fisheries and Ocean Sciences, University of Alaska Fairbanks, 17101 Pt. Lena Loop Rd., Juneau, Alaska 99801, USA
2
US Geological Survey Alaska Science Center, 3100 National Park Road, Juneau, Alaska 99801, USA
3
School of fisheries and Ocean Sciences, University of Alaska Fairbanks, PO Box 757220, University of Alaska Fairbanks,
Fairbanks, Alaska 99775, USA
4
Present address: Glacier Bay National Park and Preserve, PO Box 140, Gustavus, Alaska 99826, USA
5
Present address: 1350 Yulupa Ave. Apt. B, Santa Rosa, California 95405, USA
ABSTRACT: We characterized small-scale movement patterns and habitat of acoustic-tagged
adult (68 to 220 cm total length) female Pacific halibut during summer and fall in Glacier Bay
National Park, Alaska, a marine protected area (MPA). We used net squared displacement analysis methods to identify 2 movement states, characterize individual dispersal patterns, and relate
habitat variables to movement scales. Movement states identified for 32 of 43 halibut consisted of
(1) a non-dispersive ‘residential’ movement state (n = 27 fish), where movement was restricted to
an average movement radius of 401.3 m (95% CI 312.2−515.9 m) over a median observation
period of 58 d, and (2) a ‘dispersive’ movement state (n = 15 fish), where movements of up to 18 km
occurred over a median observation period of 27 d. Some fish (n = 10) exhibited both movement
states. Individual fish demonstrated primarily non-random dispersal patterns including home
range (n = 17), site fidelity (return to previously occupied locations following forays, n = 6), and
shifted home ranges (n = 5). However, we also observed a random dispersal pattern (n = 4) with an
estimated mean ± SE diffusion rate of 0.9 ± 0.05 km2 d−1. Home range size increased with depth
but not fish size. Home range locations were associated with heterogeneous habitat, intermediate
tidal velocities, and depths <100 m. Observations of non-dispersive movement patterns, relatively
small home ranges, and site fidelity for adult females suggest that MPAs such as Glacier Bay may
have utility for conservation of Pacific halibut broodstock.
KEY WORDS: Movement ecology · Home range · Site fidelity · Net Squared Displacement ·
Dispersal · Marine Protected Area · Flatfish · Pacific halibut
Resale or republication not permitted without written consent of the publisher
Knowledge of fish movement patterns at multiple
spatial and temporal scales can benefit the management of mobile fish species. For example, even highly
migratory species that move thousands of kilometers
are capable of philopatry at very small spatial scales
(Jorgensen et al. 2010). Thus, understanding small
scale movement patterns and habitat associations
can be an important part of achieving a holistic
understanding of stock dynamics and assessing the
potential effectiveness of spatial management techniques such as marine protected areas (MPAs) for
migratory fish species.
The Pacific halibut Hippoglossus stenolepis (hereafter referred to as ‘halibut’) is an economically, ecologically, and culturally important flatfish species in
the North Pacific Ocean. Based on observations of
*Corresponding author: [email protected]
© Inter-Research 2014 · www.int-res.com
INTRODUCTION
230
Mar Ecol Prog Ser 517: 229–250, 2014
large-scale seasonal and ontogenetic movements
during larval, juvenile, and adult life history stages
(Valero & Webster 2012), halibut in North America
are managed on a large scale (Clark & Hare 2006)
where a single stock assessment is conducted for a region that ranges from California to the Bering Sea before the allowable harvest is apportioned into smaller
management units (Webster & Stewart 2014). Some
proportion of adult halibut conduct seasonal spawning migrations from summer foraging locations in
near-shore areas to winter off-shore spawning areas
in deeper waters on the continental slope of the Pacific Ocean (Loher & Seitz 2006, Loher 2011, Seitz et
al. 2011). Recent pop-up satellite archival tagging
and conventional tagging research has demonstrated
that a large proportion of adult halibut exhibit interannual site fidelity and homing to summer foraging
locations (Loher 2008). These observations suggest
that knowledge of movement patterns at smaller
scales will be important for understanding the spatial
sub-structure of the halibut stock, potential local effects of intense fishing, and the utility or effectiveness
of MPAs as a management tool for halibut.
In addition to coast-wide, large-scale management
through area-specific harvest rates, halibut are also
regulated at smaller spatial scales through catch
sharing plans as well as the existence of MPAs. For
example, halibut harvest is restricted in the interior
waters of Glacier Bay National Park in southeastern
Alaska, where commercial fishing for halibut is being
phased out (36 CFR 13.1130-1146) over several decades and sport fishing is limited by daily vessel
quotas (36 CFR 13.1150-1160) during the summer
months. Glacier Bay National Park was added to the
National System of Marine Protected Areas in 2009.
As a large, high-latitude MPA, Glacier Bay may
eventually protect halibut that reside within its
boundaries from commercial harvest. However, obtaining information on the scale and patterns of halibut movement and habitat associations is critical for
understanding Glacier Bay’s potential effectiveness
at retention of adults (Kramer & Chapman 1999) and
specific benefits that may result from protection.
Here, we present information on the spatial and
temporal scales of movement by adult halibut in Glacier Bay National Park during summer and fall that
may be valuable for assessing the potential effectiveness of Glacier Bay National Park as an MPA. We use
net squared displacement (NSD) analysis techniques
to (1) identify and characterize 2 distinct movement
states, ‘residential’ and ‘dispersive’, (2) classify and
quantitatively describe dispersal patterns for individual tagged halibut, and (3) describe habitat associa-
tions and relationships between habitat variables
(depth, average tidal speed, habitat complexity, and
substrate type) and scale of movement for the residential movement state. We interpret these results in
terms of spatially explicit fisheries management
applications such as MPA design and effectiveness.
We conclude by addressing the potential contribution of NSD analysis methods for characterizing the
movement patterns and dispersal scales of fishes and
facilitating MPA design.
MATERIALS AND METHODS
Study area
The study was conducted in the northern portion of
southeastern Alaska within the inside waters of Glacier Bay National Park (Fig. 1). The technical boundary for the Glacier Bay National Park Marine Protected Area extends to the outside waters, but in this
document we refer to the functional MPA of the interior waters, known as ‘Glacier Bay Proper’, within
which commercial fishing and vessel traffic are regulated by the National Park Service. Glacier Bay is a
glacial fjord that is influenced by both current and
historical glacial activity. Glaciers have receded
more than 100 km in the last 300 yr in Glacier Bay,
136°0'W
137°0'W
59°0'N
Multibeam
area
Depth (m)
0
450
Core study
area
Release locations
Canada
Alaska
MPA
Boundary
Glacier Bay
Gulf of Alaska
0
58°30'N
10 km
Fig. 1. Study area within the inside waters of Glacier Bay
National Park. Map depicts the MPA boundary (red line),
multibeam survey area for habitat analyses (yellow line),
tagged fish release locations (yellow circles), and core study
area (black square) where most tagged halibut were
released and tracked
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
leaving behind a Y-shaped body of water with deep
(200 to 450 m) marine basins interspersed with shallow moraines and tidewater glaciers at the heads of
the fjords. Substantial glacial freshwater runoff influences the oceanography with high sedimentation
and areas of cold water upwelling. Strong tidal currents mix the water column completely in the shallow
lower portion of the bay, but deeper upper reaches
are largely stratified. Primary productivity levels are
highest in a transition zone in the central portion of
the bay that is characterized by intermediate stratification. Salinity, temperature, and light penetration
decrease towards the heads of the fjords (Etherington
et al. 2007b).
Fish tagging and tracking
A total of 43 halibut were captured on longlines,
tagged and released in Glacier Bay during the summers of 1991 to 1993 (Table 1, Fig. 1). Longlines were
set at 4 general release locations within the study area
using snap-on gangions designed for the commercial
halibut fishery and were ‘soaked’ for 6 h. Capture locations were determined when each fish was brought
on board the capture vessel using a PLGR GPS that
removed selective availability errors. We generally
selected larger fish (>100 cm total length, TL) for tagging because we were primarily interested in fish that
were vulnerable to the commercial fishery (≥82 cm)
and we wanted to minimize possible effects of large,
long-life acoustic tags on behavior.
Acoustic transmitters (Sonotronics) that transmitted
a unique identifying sonic pulse were attached to halibut externally during 1991 and 1992 (n = 26) and internally during 1992 and 1993 (n = 17). Externally attached acoustic tags were secured to fish by inserting
2 Teflon-coated stainless steel wires through the dorsal musculature immediately ventral to the dorsal fin,
with a backing plate of neoprene rubber and fiberglass. A sterilized needle was used to thread the wire.
For the internal attachment, tags were surgically implanted in the coelomic cavity using sterile methods.
Tags were inserted into the coelomic cavity through a
5 cm incision on the eyed-side, parallel and 2 to 3 cm
dorsal to the long axis of the fish. The incision was
closed with 7 to 8 external sutures (2-0 Braunamid
non-absorbable). During the 5 to 15 minute surgery,
the gills of the fish were irrigated with ambient seawater which was well-mixed and high-saline in the
study area. When possible, information on the sex of
the tagged fish was obtained through cannulation or
observation during surgical implantation. Tagged
231
fish were released within 500 m of the location where
they were brought on board.
Acoustic tag transmission frequency and size varied during the study. Acoustic tags attached during
the first year (n = 9) transmitted at a frequency of
80 kHz, whereas 35 kHz acoustic tags (n = 34) were
used in the 2 subsequent years due to their increased
detectability in Glacier Bay’s waters. We used 2 sizes
of acoustic tags in the study. The smaller tags (n = 17)
were 95 mm long × 18 mm diameter, weighed 16 g in
water, and had an observed lifetime of 1.3 to 2 yr. The
larger tags (n = 26) were 95 mm long × 34 mm diameter, weighed 34 g in water, and had an observed
lifetime of 2.5 to 3.4 yr. Details of tag attributes for
individual tagged fish are provided in Table S1 in
Supplement 1 at www.int-res.com/articles/suppl/
m517p229_supp.pdf.
Tagged halibut were tracked from a vessel using a
bow-mounted dual hydrophone assembly lowered
2 m beneath the surface of the water and capable of
rotating 360°. One hydrophone faced forward and
−10° from horizontal and the other hydrophone
pointed downward. These directional hydrophones
(Sonotronics DH-2) had a beam width of ± 6° and a
sensitivity of 84 dBV and were connected to manual
receivers (Sonotronics USR-4D). With this configuration, in situ range tests indicated that tags could be
detected at distances of up to 2 km. When a tag was
detected, the vessel operator maneuvered the vessel
in a circular pattern in the vicinity of the tag until signal strength was uniform at all points on the circle
and the signal received on the downward-facing
hydrophone in the middle of the circle was highly
amplified. A GPS was used to obtain the location of
the vessel at this position, which served as the estimated position of the tagged fish. Positions of tagged
fish were obtained daily to weekly during tracking
periods that lasted 3 to 6 mo, mostly in the summer
and fall, of each year. Searches for tagged fish were
conducted in an outward spiral starting from each
individual’s last known position. Consequently, if a
tagged halibut moved more than a few kilometers
away, it was not necessarily found during the subsequent search. An example of the spatial distribution
of tracking effort (number of days tracked per season) in the study area during 1991 is shown in Fig. S2
in Supplement 1.
The precision of position estimates for tagged fish
was likely to decrease with increasing water depth.
We estimated the precision of each observation
based on a linear regression of error radii vs. depth
for (1) known positions of tags recovered by SCUBA
divers (n = 3) and (2) root mean squared distances
7/24/91
7/30/91
8/20/91
8/20/91
8/20/91
8/20/91
8/20/91
9/9/91
9/9/91
6/3/92
6/17/92
6/19/92
6/19/92
6/19/92
6/23/92
7/1/92
7/1/92
7/8/92
7/8/92
7/8/92
7/8/92
7/8/92
7/8/92
7/8/92
7/8/92
7/8/92
7/31/92
11/17/92
11/17/92
11/17/92
11/17/92
6/2/93
6/18/93
6/18/93
6/18/93
6/19/93
6/19/93
6/19/93
6/19/93
6/20/93
6/20/93
6/20/93
8/11/93
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
130.6
149
132.5
168
220
154
162
151.2
140
182.8
155
151.6
152.8
149.8
164.4
77.6
157
128
94
68
76.5
153
129
104
146
127
132.5
121
124
142.5
134
125
129
201.3
136
91
136
115
140
78
90
96
117
9188
10872
1652
614
3190
17954
–
–
1170
3526
859
–
2269
1109
1296
807
3691
16658
–
4558
762
874
9291
2390
2269
2163
13393
1848
8184
237
16478
12616
12494
2014
13334
2172
1492
–
9796
1778
2031
14177
9619
207
496
497
73
1949
15963
–
–
567
2614
843
–
507
726
359
619
2027
11472
–
4558
762
378
418
2390
1497
2163
13393
213
1909
237
16478
3993
3839
1296
6369
1475
150
–
9614
984
319
14177
8023
SF
SF
U
HR
HR
R
N
N
HR
HR
HR
N
HR
HR
HR
HR
FSHR
U
N
U
U
HR
SF
R
HR
U
SHR
HR
SF
U
R
SF
SF
HR
FSHR
HR
HR
N
SHR
HR
HR
R
SHR
| | ||
|
100
June 1
| |
| | |
|
150
#
|
#
Release
Unknown
Residential
Dispersive
1993
#
/
|
|
| ||
| | ||
1992
1991
Release Length
Max
Net
Model Tracking
Date
(cm) disp. (m) disp. (m) code
year
1
2
3
4
5
6
7
8
9
Fish
ID
|
|
|
|
||
#|
#
#
#
# |
# |
#
#
#/
#/
| |
||
||
||
||
||
||
||
||
||
||
||
||
||
||
|
||
||
|
|
|
|
|
|
|
|
|
|
|
|
|
|
#/
|
|
|
|
| |
| |
| |
| |
| |
|
| |
| |
|||
|||
|||
|||
|||
|||
|||
|||
|||
|||
|||
250
Day of the year
200
||
||
||
||
|
||
|
||
||
||
||
|
||
||
||
||
||
||
||
||
||
||
|
||
||
Tagged
1992
|
|
| | | || ||
| | |
| |||
|| ||
|| ||| || ||| ||
||| ||
|| | | || ||| || |
|| ||| ||| | |
|| ||
|
# /| | || ||| |||| | | || || ||
|
| |
| | | || || || ||| | |||
||
|
|
||
|
||
|| | ||| | | || | | | | ||
||
|||
|
|| ||| | | | | |
|
#
|
#//
| ||| || | || |
|
|
Nov 1
|| |
|
|
|| ||
300
|
|| |
|
#
#|
#/ /
#/ /
||
|
|
|
|
|
| |
Jan 1
350
| |
| || || |||
| ||| |
| | ||
| || || |||||| |
| || | |
| || || |||||| ||| ||| || |
|
| ||| | ||| || ||| |||
| | ||
| |
Dec 1
||| | | | | | | | | | | | |
||| ||
|||| ||| ||| || | | | | | | || | | | | | | | | | |
||| |
|| | ||| ||| ||| | | | | | | || | | | | | | | | | | | | | | |
Oct 1
|
|
|
| | || | |||| |||
| | || | || |||
#/// |||
#
#///| |||
#//| |||
# //
Sept 1
|| ||| ||| || | | ||
| || | ||
||| || || || | || | || || ||| ||| | |||
#// |||
#///
Aug 1
| | || | | | || | | | | | ||
#
#
#/
#/
| | | ||
#
| ||
#/|
#|| |
#/ / || | | ||
#/
#/ || |
#/ / | | ||
| ||
#
#/
||
#|
#| | | | |||
#| | | | |
July 1
| |
Table 1. ID number of tagged halibut Hippoglossus stenolepis, release date (dd/mm/yy), total length, maximum horizontal displacement observed (‘max disp.’), net horizontal displacement at last observation (‘net disp.’), dispersal pattern model code (see Table 2), tracking year, date of each observation, and movement state code (see
key in table). (–) No data
232
Mar Ecol Prog Ser 517: 229–250, 2014
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
between repeated observations of motionless tags (n
= 6). The depth of each observation was multiplied by
the resulting slope coefficient, 0.65 (r2 = 0.83, p =
0.0005), to obtain error buffers for each observation
that ranged from approx. ±10 m at depths of 10 m to
approx. ±100 m at depths of 150 m.
Data analysis
Due to the large study area and the opportunistic
nature of the fish resightings, the dataset was characterized by irregular sampling intervals, unequal sample sizes among fish, and small numbers of observations for some fish. Because most movement analysis
methods require regular and frequent observations
of tagged fish, we employed an alternative analysis
framework that is robust to missing data and small
sample sizes. This analysis framework, based on the
net squared displacement (NSD) statistic, is based on
the identification of patterns of dispersal over time
that correspond to different behaviors such as foraging or migration (Börger & Fryxell 2012). NSD, also
commonly referred to as R2n, is the square of the distance between the origin of a given trajectory and
each subsequent position.
NSDt = 冷 xt – x0 冷2
(1)
where xt is the coordinate vector at time t (i.e. the
233
latitude and longitude of a fish on day t), and x0 is the
coordinate vector for the origin of the trajectory (i.e.
a fish’s release location). For random movement, e.g.
Brownian motion, the NSD statistic increases linearly
with time (Kareiva & Shigesada 1983) and the slope is
proportional to the rate of diffusion (Börger & Fryxell
2012). For non-dispersive movement, such as home
range behavior, the NSD statistic reaches a constant
value over time that represents the spatial scale of the
area in which the fish moves (Turchin 1998, Moorcroft
& Lewis 2006). For directed movement toward a specific location, such as during migration or moving between foraging locations, the relationship between
NSD and time is exponential (Nouvellet et al. 2009).
Movement states
We defined 2 different movement states using the
NSD statistic. The first movement state, ‘residential,’
reflects non-dispersive movement and was defined
when the slope of NSD vs. time = 0 (p > 0.05 for the
slope coefficient in a linear regression) for a minimum sample size of 4 consecutive observations
(Fig. 2). For this movement state, the intercept of
NSD vs. time provides information about the spatial
scale at which NSD values do not increase or decrease over time, thus providing an estimate of home
range size that is robust to small sample sizes and
B
A
6 x 107
Home range
dispersal pattern
R = residential
D = dispersive
Home range
NSD (m2)
R
Foray
0
240
C
7
6 x 10
280
300
Shifted home range
dispersal pattern
D
R
R
320
340
R
D
0
Fish ID
3
240
260
280
300
320
340
Day of the year
5
Release location
3
5
Depth (m)
-250
-200
-150
-100
-50
260
Shifted home range
0
2 km
Fig. 2. Hippoglossus stenolepis. (A) Different dispersal patterns as shown by 2 tagged halibut (ID#3 and ID#5). The size
of solid red and yellow circles represents estimated telemetry position error (in m). Dotted elipses: home ranges. (B,C)
Corresponding plots of the net squared displacement (NSD)
statistic vs. time for (B) a home range dispersal pattern with
all observations classified as residential movement (ID#3),
and (C) a shifted home range dispersal pattern with observations classified as residential or dispersive (dashed and
solid-line elipses, respectively) movement states (ID#5)
Mar Ecol Prog Ser 517: 229–250, 2014
234
infrequent observations (Moorcroft & Lewis 2006).
Consecutive observations classified as residential are
subsequently referred to as ‘home ranges’. We estimated typical home range size for the residential
movement state using an intercept-only linear
mixed-effects model:
log(NSD)ij = α + ai + εij
ai ~ N(0,σ2a), εij ~ N(0, σ2)
(2)
where α is the fixed-effects estimate of mean home
range size for the population of i home ranges, ai is a
random variable that represents the variation of individual home range estimates around the fixed-effects
mean, and within-group error εij is assumed to be
independent and normally distributed with a mean of
zero. The model was fit using restricted maximum
likelihood. NSD values were log-transformed to account for heteroscedasticity prior to modeling. We
applied the bias-correction for a log-normal distribution to back-transform the estimated mean to the
original scale:
Home range size (NSD) = exp(α + s2/2)
(3)
where α is the mean home range size for the population as described above and s2 is the estimated variance for α. To provide a more intuitive linear description of movement scale, the square root of intercept
coefficients were reported as a ‘home range radius.’
To minimize potential bias of capture and tagging on
the scale of home range movement patterns (e.g.
temporary tagging effects, uncertainty in longline
position during capture vs. release location, or travel
from release position back to the home range), observations within 3 d of tagging were not used for home
range analyses.
The second movement state, ‘dispersive,’ was classified as all observations where the slope of NSD vs.
time ≠ 0 (Fig. 2C). This movement state generally
contained observations from both random and directed movement types. Because fish with more mobile
movement patterns were more difficult to relocate,
observations were not collected frequently enough to
determine whether individual observation sequences
were random (linear relationship for NSD vs. time) or
directed (exponential relationship for NSD vs. time).
Therefore, we did not summarize this movement
state using a mixed-effects model, as we did for the
residential movement state, because it likely contained a mix of both movement types.
Although we were not able to explicitly classify observations as either random or directed, insight into
the randomness of the dispersive movement state
was obtained by comparing our observations of average NSD vs. time to expected random values using
correlated random walks (CRWs). CRWs are movement paths comprised of a discrete series of movement steps where the expected distance and the
expected angle between subsequent steps determines overall movement path characteristics such as
diffusion rates (larger for larger step lengths) and
directed vs. random movement (more directed for
smaller variation in turning angles, more random for
larger variation in turning angles). CRWs are usually
simulated based on empirical distributions of both
step lengths and turning angles obtained from frequent and regular observations of a fish’s location
over time (Kareiva & Shigesada 1983, Turchin 1998).
However, because our dataset was highly irregular,
we simulated random movement from a step length
distribution comprised of all records that were 1 d
apart, but used a theoretical distribution of turning
angles (wrapped Cauchy with an autocorrelation
coefficient of 0.5) that is typically observed during
fish foraging activities (Morales et al. 2004, Bartumeus et al. 2005). This approach allowed us to
incorporate empirical information on spatial scales of
our tagged fish (daily step lengths) under a specific
hypothesis of random movement during foraging. We
fit exponential curves (Moorcroft & Lewis 2006) to
both the residential and dispersive step length distributions and sampled randomly from these distributions to create 1000 CRWs for a duration of 90 d for
each movement state. To account for the effects of
movement that occurs within the confined waters of
the study area, simulations were conducted on a 20 m
bathymetry grid of Glacier Bay (Geiselman et al.
1997) and were initiated at the first location of each
observed residential or dispersive movement sequence. If a simulated position for the CRW fell on
land, that position was discarded and a new coordinate was chosen. To assess randomness, the mean
NSD from observed data was compared to the mean
and 95% CI from the CRWs at each time step. NSD
values for random movement should fall within the
95% CI for the CRW simulations, whereas values for
directed and non-dispersive movement will be
greater than the upper bound and less than the lower
bound, respectively, of the 95% CI for CRW values
(Austin et al. 2004).
Individual dispersal patterns
In addition to characterization of movement states,
knowledge of the way in which NSD changes over
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
time during non-dispersive, directed, or random
movement can be used to formulate theoretical
models that describe behavioral phenomena such as
home range occupation, directed movements to new
locations, migrations or forays, or random movement
(Bunnefeld et al. 2011, Börger & Fryxell 2012, Papworth et al. 2012, Singh et al. 2012). We identified 5
models of NSD vs. time, subsequently referred to as
dispersal patterns, that we assume represent underlying behaviors for tagged fish in this study
(Table 2). (1) The ‘home range’ dispersal pattern
(HR) reflects a restricted range of movements described by a slope of zero for NSD vs. time. (2) The
‘random’ dispersal pattern (R) reflects diffusion (e.g.
Brownian motion) and can be described by a linear
increase in NSD vs. time. (3) The ‘shifted home
range’ dispersal pattern (SHR) represents movement
from the release location to another location in the
study area and consists of a non-linear (sigmoidal)
model with parameters for the timing of the midpoint of the travel to the new location (θ, day of the
year), a scale parameter (ϕ, day of the year) to estimate the time to travel between the midpoint and
235
approximately ¾ of the distance to the destination
(Bunnefeld et al. 2011), and the squared distance to
the destination (δ, m2). (4) The ‘site fidelity’ (SF) dispersal pattern represents departure from the release
location and subsequent return to the original location. It consists of a double sigmoidal model, with
one sigmoid function to describe the migration start
(subscript m) and one to describe the return to the
original location (subscript r). The SF dispersal pattern has the same parameters as the SHR model, but
with an additional parameter, θr, to describe the timing (day of the year) of the midpoint for the return.
(5) The ‘foray/shifted home range’ dispersal pattern
(FSHR) reflects departure from the release location
and dispersive behavior before resuming HR behavior in another location. The parameters for this
model are the same as those for the SF pattern, but
an additional parameter (δr, m2) is added to describe
the squared distance between the farthest distance
traveled during the foray and the SHR.
Dispersal patterns for individual tagged fish were
classified by determining which of the 5 models for
NSD vs. time described above best described the
Table 2. Model name, code, number of parameters, theoretical example, and formula for dispersal patterns used to classify
movement trajectories for halibut Hippoglossus stenolepis. t is time, ∝ is the intercept of a linear model, β is the slope of a linear
model, and δ, θ, and ϕ are nonlinear model coefficients that describe the spatial extent, timing, and temporal scale respectively. Subscripts — ‘m’: migration start; ‘r’: return. See Bunnefeld et al. (2011)
Model (code)
No.
params
1. Home range
(HR)
1a
2. Random (R)
1
3. Shifted home
range (SHR)
3
4. Site fidelity (SF)
4
5. Foray/shifted
home range
(FSHR)
6
Ecologicalinterpretation
NSD vs. time
Formula
Constrained movement,
such as home range
behavior (residential
movement state)
NSD = ∝
Nomadic, diffusive
movement such
as foraging
NSD = βt
Movement to new
location
NSD =
Migration or foray
with return to
original location
NSD =
Migration or foray
with return to
new location
NSD =
δ
⎛θ−
1 + exp ⎜⎜
⎝ ϕ
t⎞
⎟⎟
⎠
δ
⎛ θm − t ⎞
1 + exp ⎜⎜
⎟
ϕ ⎟⎠
⎝
δm
⎛ θm −
1 + exp ⎜
⎜⎝ ϕ m
t⎞
⎟
⎟⎠
+
+
−δ
⎛ θr −
1 + exp ⎜⎜
⎝ ϕ
t⎞
⎟⎟
⎠
− δr
⎛ θr −
1 + exp ⎜
⎜⎝ ϕ r
t⎞
⎟
⎟⎠
Time
a
Home range can be modeled as intercept-only when the sampling interval is greater than the time the fish takes to reach the
limits of its range
Mar Ecol Prog Ser 517: 229–250, 2014
236
observed values of NSD vs. time based on model
selection techniques. Non-linear models (all models
besides HR and random) were fit using a non-linear
least squares algorithm (the nls function in the ‘stat’
package for R). The best-fitting model for each individual trajectory was selected using the Akaike
Information Criterion adjusted for small sample
sizes (AICc) (Burnham & Anderson 1990) and residual analysis. An example of the dispersal pattern
classification process for individual fish is provided
in Fig. S3 in Supplement 2 at www.int-res.com/
articles/suppl/m517p229_supp.pdf. Once individual
fish were classified according to dispersal pattern,
we calculated the average for: (1) maximum distance from the release location during the observation period, (2) distance from the release location at
the end of the observation period, (3) observation
period duration, and (4) fish size for each dispersal
pattern.
We used mixed-effects models to summarize model
parameters for dispersal patterns to which more than
3 fish were assigned. Because fish with the HR dispersal pattern were included in the mixed-effects
model for the residential movement state, mixedeffects models were only used to summarize the random and SF dispersal patterns. For the random dispersal pattern, we quantified the rate of dispersal
over time using a linear model with no intercept, as,
by definition, dispersal must be zero at the origin of a
trajectory, using fish ID as a grouping variable.
NSDij = βt + bi t + εij
bi ~ N(0,σ2b), εij ~ N(0, σ2)
(4)
where β is the fixed-effect variable estimate of the
mean slope of NSD vs. time for the population of i
random trajectories, bi is a random variable that represents the variation of individual slopes around the
population mean slope, and within-group error εij is a
random variable that is independent and normally
distributed with a mean of 0. The model was fit using
restricted maximum likelihood. Because random
movement results in the process of diffusion, NSD vs.
time for random movement is proportional to diffusion. Therefore, results are presented in the form of
the estimated rate of diffusion, in km2 d−1, which is
calculated by dividing the slope of NSD vs. time by 4
for movement in 2 dimensions (Börger & Fryxell
2012).
For the SF dispersal pattern, we quantified timing,
duration, and distance traveled during forays that
occurred during the summer with a non-linear
mixed-effects model:
NSDij =
δ + di
−δ + di
+
+ εij
(θ + r ) − t
θm − t
1 + exp
1 + exp r i
ϕ + fi
ϕ + fi
(
)
(
)
(5)
di ~N(0,σd2 ), fi ~ N(0,σ 2f ), ri ~ N(0,σr2), εij ~N(0,σ 2 )
where δ, θm , θr, and ϕ are fixed-effects parameters
for the asymptote (e.g. migration distance), date of
migration, date of return, and scale, respectively (see
Table 2), and di , fi , and ri are random-effects variables assumed to be normally distributed with mean
0 that represent individual variation in the asymptote, scale, and date of return, respectively. Withingroup error εij is assumed to be independent and normally distributed. The estimate for distance traveled
during the foray is reported as the square root of the
asymptote, δ. As approx. 95% of the distance between the midpoint of the migration and arrival at
the new location occurs over the time span of 3 × ϕ
(Börger & Fryxell 2012), timing of migration is estimated by θm − (3 × ϕ) and timing of return by θr + (3 ×
ϕ). Population estimates of average foray duration
are calculated as the difference between the two:
[θr + (3 × ϕ)] − [θm − (3 × ϕ)].
Selection of random-effects variables for the SF
model was conducted by first examining the range of
coefficient values for each parameter, based on separate fits of the model to each trajectory, and selecting
parameters with large variation as random effects in
the full model (Pinheiro & Bates 2000). Alternative
models with fewer random-effects variables and
autocorrelation structures were tested against the
null model using maximum likelihood and compared
using AIC and likelihood ratio tests. The best model
(Eq. 5) also included an AR1 autocorrelation coefficient of 0.3. All mixed-effects models assumed a
Gaussian error structure and were fit using the
library ‘nlme’ in the R program. The assumption of a
normal distribution in random-effects estimates was
checked using the Shapiro-Wilks test for normality.
Habitat relationships
We characterized habitat occupied by tagged halibut during the residential movement state using
several habitat metrics available for the study area
(depth, slope, habitat complexity, rugosity, substrate
type, tidal velocity). Because > 90% of the tagged fish
observations occurred within a large area of the central portion of the bay that was characterized by a
multibeam survey in 2001, we were able to use finescale depth (Carlson et al. 2002) and habitat information (Harney et al. 2006) resulting from this survey.
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
However, observations for 2 tagged fish were removed from habitat analyses because a majority of
their observations fell outside of the multibeam study
area. Continuous rasters for slope, change-in-slope
(an indicator of slope interfaces and measure of habitat complexity), and rugosity (a measure of surface
roughness) were derived from 5 m resolution depth
data using ArcGIS 10.0 Spatial Analyst and ArcGIS
10.1 Benthic Terrain Modeler (Wright et al. 2012,
ESRI 2011). Continuous rasters for soft sediment and
moderate habitat complexity were derived from discrete habitat map polygons by calculating Euclidean
distance from each grid cell to each type of polygon.
Continuous information on time and depth averaged
(monthly) tidal velocity was available from a 2dimensional circulation model (ADCIRC) of Glacier
Bay (Etherington et al. 2007b, Hill et al. 2009). We
used information from these 7 continuous rasters
(depth, slope, change-in-slope, rugosity, distance from
soft sediment, distance from moderate complexity
habitat, and tidal velocity) to identify habitat associations and quantify the effects of habitat variables on
HR size. Study area maps and additional details on
habitat raster characteristics are available in Supplement 3 (Fig. S4, Table S3) at www.int-res.com/articles/
suppl/m517p229_supp.pdf.
Habitat associations. To provide a simple description of the predominant habitat characteristics observed for the residential movement state relative to
all available habitat types in the study area, we adapted an approach used to detect habitat associations
based on the spatial distribution of catch during trawl
surveys (Perry & Smith 1994). This method involves
comparing the cumulative distribution function (CDF)
of habitat values (e.g. depth) where tagged fish were
observed to the CDF of available depths in the study
area. Because halibut are large-bodied fish capable
of a high degree of movement, we assumed they
could have moved anywhere in the study area over
the course of the observation period. To obtain CDFs
for available habitat in the study area, a 20 m grid of
the study area was created in ArcGIS (1.08 × 106
points) and values from each habitat raster were
extracted at each grid point.
To account for telemetry error in the habitat analyses, a buffer with a radius of the estimated error
was drawn around each tagged fish observation and
all grid values within the buffer were averaged. ‘Observed’ CDFs were then calculated using the median
value of all observations in each HR to avoid pseudoreplication from treating repeated, irregular observations of 1 fish at 1 location as independent events
(Rogers & White 2007). Confidence intervals for
237
observed CDFs were generated by bootstrapping,
where the median observation for each HR was sampled with replacement 1000 times, and the 0.975 and
the 0.025 quantile values were selected as the upper
and lower confidence intervals.
We defined habitat associations by quantitatively
comparing the CDFs for observed and available fish
habitat. Specifically, for each habitat variable, we
used the bootstrapped confidence levels for the observed CDFs to test for differences between observed and available habitat using the KolmogorovSmirnov (K-S) test. The K-S test is frequently used to
test for differences between CDFs based on the maximum vertical difference (D) between the CDFs
(Conover 1999). To determine whether positive differences existed between the observed and available
CDFs (e.g. an association with shallower depths), we
found the greatest positive difference (D+) between
the upper CI of the observed CDF and the available
CDF. To determine whether negative differences existed between the observed and available CDFs (e.g.
an association with deeper depths), we found the
greatest negative difference (D−) between the lower
CI of the observed CDF and the available CDF. We
determined D+, D−, and p-values for each habitat
variable using one-tailed Kolmogorov-Smirnov tests.
Habitat and home range size. We used a generalized additive model (GAM) to determine whether HR
sizes were related to habitat variables or fish total
length. Intercept coefficients from the mixed-effects
model for the residential movement state (in log format) were used as the response variable. Explanatory habitat variables were selected from the 7 continuous habitat rasters used for habitat association
analyses. In addition to habitat variables, we also
included fish total length and year of study as explanatory variables for HR size. The GAM approach
was used to allow for potential non-linearities in the
relationship between response and explanatory variables. Prior to analysis, all variables were checked for
covariance with the Pearson correlation coefficient; if
a set of variables were found to be correlated, only
the variable with the strongest relationship with the
response variable was used in the model. After assessing correlation and linearity of habitat variables,
2 full models were tested:
Model 1: y = α + s1 (depth, k = 2) + s2 (fish total length,
k = 3) + s3 (distance from moderate complexity, k = 3)
+ β1 change-in-slope + β2 tidal velocity + year + ε (6)
Model 2: y = α + s1 (depth, fish total length, k = 3)
+ s3 (distance from moderate complexity, k = 3)
+ β1 change-in-slope + β2 tidal velocity + year + ε (7)
Mar Ecol Prog Ser 517: 229–250, 2014
238
where y is the vector of estimated intercepts from the
mixed-effects model for all residential trajectory segments (n = 29), α and βi are regression coefficients, si
are smooth functions of the predictor variables, k
represents the degree of smoothing in the smooth
functions, and ε are the residuals, assumed to be
independent and normally distributed. GAM models
were fit using maximum likelihood methods with a
Gaussian error structure in the mgcv package in R
(Wood 2006). Variables were sequentially removed
from a full model based on the highest p-value (i.e.
larger than 0.05) and the best model was chosen
based on the AICc criterion and residual analysis.
and the location of the last observation was 3.5 ±
0.8 km. The average (± SE) maximum distance traveled during the entire observation was 5.8 ± 0.9 km.
The maximum distance from release location recorded during the study was 17.9 km (Table 1).
There were no significant relationships between the
maximum distance traveled for each fish and fish
size (linear regression, p = 0.709), tag size: body
weight ratio (linear regression, p = 0.637), tag size
(small vs. large; ANOVA, p = 0.146), or tag attachment method (interval vs. external; ANOVA, p =
0.797).
Movement states
RESULTS
Fish tagging and tracking
A total of 43 fish were tagged between 1991 and
1993 (Table 1). Most fish were tagged between June
and September of each year, but 4 fish were tagged
in November 1992 and tracked during the following
summer, and all were released in good condition.
Tagged fish TL (mean ± SD) was 133 ± 32 cm. Almost
all (16 of 18) of the fish that we were able to sex were
female; however, sex could not be determined for
the majority (n = 25) of tagged halibut in this study.
Based on fish size and maturity ogives from International Pacific Halibut Commission (IPHC) records
during this time period, the majority of the halibut
tagged in this study were likely to be adult females
(T. Loher pers. comm.).
Five fish were never relocated following tagging
(Table 1). For the remaining 38 fish, the mean (± SD)
number of relocations per fish was 17.4 ± 14.3 and
ranged from 1 to 49. More than half of the relocations
for individual fish were obtained within 3 d of the
previous observation, and 90% of the subsequent
observations in each tracking period were within 8 d
of the previous observation. Thus, the temporal scale
of tagged fish observations during each tracking
period can be characterized as daily to weekly. In
total, 706 acoustic tracking position estimates were
obtained for all tagged fish in all years. Tracking
effort differed among years, with most intense tracking during 1991 (32.5 observations per fish) and
decreasing during 1992 (13.4 observations per fish)
and 1993 (14.8 observations per fish). Tagged fish
were observed over a mean tracking duration of
79.5 d (range 1 to 290 d) each year.
The average (± SE) distance that individual tagged
fish (n = 38) moved between the release location
The residential movement state was observed most
frequently (27 of 43 tagged halibut; Fig. 3A). A total
of 31 residential movement sequences (some fish had
more than 1 residential sequence) were observed
with a median duration of 58 d. The mixed-effects
model population estimate (mean ± SE) for the intercept of NSD vs. time was 12.0 ± 0.3 m2, with a standard deviation for random effects of 1.4 on the log
scale (Fig. 4A). This corresponds to an estimated
population HR radius of 401.3 m (95% CI = 312.2−
515.9 m) and 95% CIs for individual HR radii that
range from 104.3 to 1493.9 m on the untransformed
scale.
The dispersive movement state was observed for
15 of 43 tagged halibut (Table 1, Fig. 3B). A total of 18
dispersive movement sequences were observed with
a median duration of 27 d. This duration was significantly shorter than that of the residential movement
state (t-test, p < 0.0001). The average maximum distance from the release location for fish that exhibited
the dispersive movement state was 10.9 km.
The step length distribution for the residential
movement state (Fig. 5A) was significantly different
than the step length distribution for the dispersive
movement (Fig. 5B). Based on a randomization test
with 1000 permutations, the median daily movement
step length for observations from the residential
movement state (330.4 m, n = 193 observations) was
significantly less (p < 0.0001) than the median daily
movement step length for observations from the dispersive movement state (861 m, n = 19 observations).
The rate parameter for the exponential curve that
was fit to each step length distribution for use in the
CRW analyses was 0.000213 ± 0.000015 (SE) for
observations from the residential movement state
and 0.0008059 ± 0.000184 for observations from the
dispersive movement state.
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
log(NSD) (m2)
16
239
A
14
12
10
8
100
150
200
250
300
350
Day of the year
NSD (m2)
B
108
0
180
200
220
240
260
280
300
Day of the year
4 x 108
NSD (m2)
C
2 x 108
0
Fig. 3. Hippoglossus stenolepis. Movement patterns of tagged
halibut exhibiting (A) residential movement (colored lines:
movement sequences; black arrows: shifted home ranges
from release locations) and (B) dispersive movement (colored
arrows) within the core study area
The CRW simulations highlighted major differences in the scale and nature of dispersal between
the residential and dispersive movement patterns.
Average values of NSD observed for the dispersive
movement state were generally within the 95% CI
for the CRWs, however these tended to be closer to
0
20
40
60
80
Time (d)
Fig. 4. Hippoglossus stenolepis. Observed (points) and estimated (lines) of the net squared displacement (NSD) statistic
from mixed-effects models for (A) the residential movement
state (n = 31 home ranges) on log-scale, (B) the site fidelity
dispersal pattern (n = 4 ind., as 2 ind. with winter observations were not included), and (C) the random dispersal pattern (n = 4 ind.). Population (fixed-effect) means are shown
with thick dashed lines, 95% CIs are shown as gray polygons, and thin black lines represent individual (randomeffects) estimates (in A, the length of these lines represents
the period observation for each individual)
Mar Ecol Prog Ser 517: 229–250, 2014
240
6 x 108
A
0.0020
A
0.0015
4 x 108
0.0010
2 x 108
0.0000
0
500
1000
1500
2000
2500
8 x 10–4
3000
B
–4
6 x 10
Average NSD (m2)
Probability density
0.0005
0
0
1 x 10
20
40
60
80
7
B
8 x 106
4 x 10–4
6 x 106
2 x 10–4
4 x 106
0
2 x 106
0
1000
2000
3000
4000
5000
6000
Daily step lengths (m)
Fig. 5. Hippoglossus stenolepis. Daily step length distributions for (A) residential and (B) dispersive movement states
along with exponential curves fitted to each distribution
the upper confidence level for the first 20 d (Fig. 6A).
Like the CRWs, observed values of NSD for the dispersive movement state exhibited a general trend for
increased NSD values over time. In contrast, observed values of NSD vs. time for residential movement pattern were located along the lower 95% CI
for the simulations, and some observed values were
smaller than the CRW confidence intervals after ca.
30 d (Fig. 6B).
Individual dispersal patterns
Of the 38 fish that were relocated at least once following release, 32 produced a sufficient number of
observations to allow classification of their dispersal
pattern. More than half of these fish (n = 17) remained in the vicinity of the release location and
demonstrated an HR dispersal pattern, where the
0
0
20
40
60
80
Time (d)
Fig. 6. Hippoglossus stenolepis. Observed mean net squared
displacement (NSD) statistic (circles) compared to the mean
(dotted line) and 95% CIs (red lines) of 1000 correlated random walk simulations for (A) dispersive and (B) residential
movement states. Note the difference in scales for the y-axes
average net displacement over average tracking
durations of > 3 mo was <1 km (Table 3). Five fish
demonstrated SHR dispersal patterns (SHR, n = 3 and
FSHR, n = 2) with average net displacements of 10
and 4 km, respectively, over similar time periods. Six
fish demonstrated the SF dispersal pattern by moving
an average maximum displacement of approx.
10 km, but eventually returning to locations that
were an average net displacement of < 2 km from the
release location over average time periods of > 4 mo.
Two of these fish established HRs at other locations
in the study area during the foray. Three fish classified with the SF dispersal pattern returned to within
hundreds of meters (range 200 to 500 m) of their
release locations after moving an average (± SE)
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
241
maximum distance of 9.8 km ±
Table 3. Number of individuals of halibut Hippoglossus stenolepis assigned to
each dispersal pattern (see Table 2; HR: home range; SHR: shifted home
0.5 km. One fish classified as SHR
range; FSHR: foray/shifted home range; SF: site fidelity; R: random; U: un(ID#5, Fig. 2) also demonstrated a SF
known; N: not relocated after release) along with average (± SE) maximum
pattern during a temporary foray of
horizontal displacement, average net displacement at the end of the ob6 km and duration of 16 d followed by
servation period, average observation period duration for each pattern, and
total length. (–) No data
a return to within 200 m of the location occupied prior to the foray. Four
Model
No.
Max disNet disObs.
Total
fish exhibited the random (R) disperof
placement
placement
duration
length
sal pattern, moving an average maxifish
(km)
(km)
(d)
(cm)
mum and average net displacement
of approx. 12 km over observation
HR
17
1.7 (0.02)
0.9 (0.2)
102.5 (14.3)
142.7 (10.0)
periods that averaged < 2 mo (approxSHR
3
10.9 (1.2)
10.3 (1.6)
101.7 (26.4)
129.8 (6.8)
FSHR
2
8.5 (4.8)
4.2 (2.2)
110.0 (8.0)
146.5 (10.5)
imately half of the typical durations
SF
6
10.4
(0.8)
1.8
(0.7)
146.5
(46.5)
131.1
(3.7)
observed for the fish assigned to other
R
4
12.8 (3.5)
12.3 (3.3)
48.3 (14.8)
122.0 (13.4)
dispersal patterns). The 6 fish for
U
6
4.3 (2.5)
3.3 (1.8)
13.5 (11.7)
134.8 (12.9)
which only a few observations were
N
5
–
–
–
112.4 (12.9)
collected (U) had very short observation durations (average 13.5 d), yet
the average net displacement of
Table 4. Habitat associations and habitat occupancy ranges for the residential
approx. 3 km for these fish was
movement state of halibut Hippoglossus stenolepis. Bold: significant Kolmogreater than that observed for the HR
gorov-Smirnov test statistics (D). One-tail test sign: positive: the upper 95% CI
dispersal pattern. There was no
for the observed CDF is greater than the available CDF; negative: the lower
95% CI for the observed CDF is less than the available CDF). Habitat value given
significant difference among the total
only where significant D+ or D− statistics were obtained. NS: not significant
lengths of fish in each of the 5 dispersal patterns, unclassified movements,
Habitat variable
D
p
1-tail Habitat
or the fish that were never observed
test
value
after tagging (Kruskal-Wallis test, p =
sign
at D
0.3679, df = 6).
A mixed-effects model was fit to
Depth (m)
0.1701
0.003
+
104
Slope (°)
0.2674 < 0.001
−
1.0
NSD data from 4 of 6 tagged fish that
Change in slope (°)
0.3234 < 0.001
−
3.6
were classified as having a SF disperRugosity (ratio)
0.2066 < 0.001
−
0.00015
sal pattern (Fig. 4B). Movements for
Tide speed (m s−1)
0.2114 < 0.001
−
0.07
the 2 fish that were not included ocTide speed (m s−1)
0.2189 < 0.001
+
0.34
curred over the winter, so the timing
Distance to soft bottom (m)
0.0127
0.968
+
NS
Distance to moderate complexity (m) 0.0785
0.288
+
NS
of their movements could not be compared to fish that were observed only
during the summer. For the 4 remainThe population (fixed effect) estimate of the slope
ing fish in this group, the fixed effect estimate for the
of NSD vs. time for the random dispersal pattern was
distance traveled during their forays (the asymptote,
3.6 ± 0.2 km d−1 (SE) (p < 0.0001) which corresponds
δ) was a movement radius of 10.5 ± 4.0 km (SE) (p <
to an estimated diffusion constant D of 0.9 ± 0.05 km2
0.0001) (Fig. 4B). The SD for individual fish from the
d−1 (SE) (Fig. 4C). In contrast to the HR dispersal patpopulation mean (di) was 5.1 km. The scale parameter
ϕ was estimated to be 2.4 ± 0.5 d (SE) (p < 0.0001), and
tern, there was very little difference between the
the SD for the corresponding random-effects variable
individual estimated movement rates because most
fi was 0.0001 d. The population estimate for the
of the variation was attributed to residual variation
timing of departure was July 8 ± 1 d (SE) (p < 0.0001),
around the fixed effect.
and the estimate for the timing of the return to the
original location was August 17 ± 5 d (SE) (p < 0.0001)
Habitat relationships
with a standard deviation for the random-effects variable ri of 10.1 d. The overall population estimate for
Habitat associations. Significant habitat associaduration of the forays was 40 ± 5.1 d (SE). The value
tions were observed between tagged fish in the
of the AR(1) autocorrelation coefficient (referred to in
residential movement state and available habitat
the nlme package as Phi) for the model was 0.61.
242
Mar Ecol Prog Ser 517: 229–250, 2014
in the study area (Table 4, Fig. 7). Relationships
were strongest for the habitat heterogeneity variables of change-in-slope, slope, and rugosity, with
tagged fish tending to occupy areas of higher
habitat heterogeneity relative to the range of values available in the study area. HRs were also
associated with intermediate values of tidal velocity, where significant differences between observed
and available habitat occurred at both high and
low values of tidal velocities. Finally, HRs were
associated with shallower depths relative to the
range of available depths in the study area, with
approximately 75% of HRs occurring in depths
less than 100 m. No significant differences were
observed between tagged fish and distance to
moderate complexity or distance to soft substrate
variables.
Habitat and home range size. The results from
the GAM analysis indicate that of the variables
examined, HR size varied most strongly with
depth. The best model contained only a depth
term with an estimated degrees of freedom of 1.8
(p = 0.007). For depths less than ~150 m, HR size
increased with increasing depth (Fig. 8). The deviance explained by the selected model was
30.0%. No other models with all significant terms
were within ± 2 ΔAICc of the best model. No patterns were observed in the residuals, which were
consistent with a normal distribution (Shapiro-Wilk
test, p = 0.65).
15
log (NSD) (m2)
14
13
12
11
10
9
Fig. 7. Hippoglossus stenolepis. Observed cumulative distribution functions (CDFs) for the residential movement state
(gray polygons: 95% CIs) compared to CDFs of available
habitat values within the study area (thick black lines) for
depth, slope, change-in-slope, rugosity, tidal velocity, distance to moderately complex habitat, and distance to soft
substrate habitat. See Table 4 for habitat association test
statistics
0
50
100
150
200
250
300
Depth (m)
Fig. 8. Hippoglossus stenolepis. Predicted relationship between depth and home range size from the best generalized
additive model (solid black line); 95% CIs are shown with
dashed lines. NSD: net squared displacement
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
DISCUSSION
Although halibut are large-bodied fish capable
of moving thousands of kilometers during winter
spawning migrations (Skud 1977, Loher & Seitz
2006), our results suggests that limited dispersion at
very small spatial scales may be a common phenomenon for adult female halibut in Glacier Bay during
the summer and into the fall. The residential movement state was demonstrated by the majority (27 of
43) of the fish tagged in this study. The HR dispersal
pattern (which consists of residential movement in
the vicinity of the release location throughout the
observation period) was also the most frequently
observed dispersal pattern (n = 17) among the 32 ind.
for which dispersal patterns could be determined.
Although fish that exhibited the dispersive movement state moved more broadly around the study
area, these movements were still relatively small
(< 20 km) compared to the distances moved during
winter migrations.
Fish that were never relocated or were relocated
too infrequently to characterize their movement patterns (11 of 43 fish) may have exhibited a more
mobile movement pattern and thus moved out of the
study area quickly. In this case, they could have
moved to areas within Glacier Bay that were not
monitored during acoustic surveys or they could
have left the interior waters of Glacier Bay entirely.
Alternatively, they may have been captured in commercial harvests that were occurring in Glacier Bay,
experienced mortality, or the tag could have been
shed or ceased to function.
Movement states
Telemetry records often document different behaviors among individuals that are driven by different
movement ‘states’ (Blackwell 1997, Morales et al.
2004). For example, a period of intensive foraging
may result in a movement state with little net displacement, while a period of migration may result in
a movement state with relatively large net displacement. Typically, ecologists are interested in the spatial and temporal scales of these movement states, as
well as habitat attributes with which they may be
associated (Papworth et al. 2012).
The 2 movement states, residential and dispersive,
that tagged fish exhibited during the summer and fall
differed in terms of scale, duration, and potential for
dispersion. The residential movement state was associated with average movement scales of <1 km for
243
several months at a time and a sustained, nonrandom lack of dispersion. In contrast, the dispersive
movement state was characterized by greater spatial
scales (approx. 10 km), shorter temporal durations
(<1 mo), and likely contained a mix of random and
directed movement.
Because tracking occurred during the summer foraging season, and large adult halibut have few predators, these 2 movement states could reflect different
underlying foraging strategies. Both ‘sit-and-wait’
ambush and active searching are common foraging
tactics for flatfish species (Gibson 2005). The residential movement pattern could be driven by a sit-andwait tactic, which would require little movement in
areas where prey is delivered to the fish. Based on
laboratory studies, a closely related congener Atlantic halibut Hippoglossus hippoglossus is thought to
be an ambush predator that employs a sit-and-wait
feeding tactic (Haaker 1975, Nilsson et al. 2010).
Other flatfish such as summer flounder Paralichthys
dentatus have been observed to employ a variety of
foraging tactics — including ambush and active pursuit — that change with prey type (Staudinger &
Juanes 2010). Thus, switching between the 2 movement states may occur in conjunction with changes in
the type, abundance, distribution, and mobility of
prey species (see Nakano et al. 1999). However, the
dispersive movement pattern could also include fish
that are moving in a directed manner from one feeding location to another.
Caveats. It is important to emphasize that the data
presented in this study are inherently positive and
biased toward the observation of the residential
movement state. The experimental design employed
in this study, which featured searching for tagged
fish in the vicinity of their last known location,
resulted in a much better characterization of the residential movement state compared to the dispersive
movement state. The tracking procedure was effective for locating tagged fish that were occupying
HRs, as the detection range for the acoustic tags (up
to 2 km) was larger than the scale of most HRs. However due to the difficulty of tracking more mobile fish
for long time periods in the large study area, it is
likely that the dispersive movement state occurred
more frequently than was observed and its spatial
extent was not fully characterized. Although detection ability was adequate for characterizing the residential state throughout the study, changes in tag
size and frequency (Table S1) could have improved
the detection of fish in the dispersive movement state
as the study progressed. It is likely that the largest
movement observed (18 km) probably reflects a prac-
244
Mar Ecol Prog Ser 517: 229–250, 2014
tical limit for the area searched during this study, so
movements beyond that would have a low probability of detection.
Although it is possible that unknown tagging effects may have affected the behavior of tagged halibut, we feel that tagging effects are unlikely to have
affected the scale and nature of halibut movement
reported in this study for several reasons. (1) A longterm laboratory study of both internal and external
archival tag attachment suggests that both types of
attachment are well-tolerated by halibut and do not
result in changes in behavior compared to controls
(Loher & Rensmeyer 2011). (2) The tags were small
relative to the size of the fish (average = 0.1%, maximum = 0.4%). (3) Pacific halibut fitted with much
larger pop-up satellite tags have been observed to
move more than 1000 km (Loher & Seitz 2006). (4) We
found no statistical relationships between fish size or
tag:body size ratio and maximum displacement, and
no relationship between maximum displacement and
tag size (small or large) or type of attachment (internal or external).
Individual dispersal patterns
Non-random dispersal patterns: home range and
site fidelity. The majority of tagged fish in this study
exhibited distinctly non-random individual dispersal
patterns that were dominated by HR, but also included temporary long-distance forays followed by
return to previously occupied locations and shifting
of HRs to new locations. The prominence of the HR
dispersal pattern suggests that regular use of relatively small areas could be a common phenomenon
during summer. Several acoustic telemetry studies
have demonstrated summer HR behavior for other
flatfish species such as adult English sole Parophrys
vetulus (S. O’Neill pers. comm.) and juvenile California halibut Paralichthys californicus (Espasandin
2012) that occurs at scales <1 km. Because some halibut shifted locations for HR behavior, it is possible
that some fish may switch HR locations depending on
changes in prey distribution and abundance and thus
may not have fidelity to specific locations.
However, multiple observations of tagged fish
returning to within several hundred meters of previously occupied locations following larger-scale
movements (e.g. 10 km distance, 1 mo duration) suggests that some halibut do have SF to specific locations (as defined by Giuggioli & Bartumeus 2012).
The SF dispersal pattern was observed for 7 of 43 fish
(including 1 fish, ID#5, that was assigned to the SHR
dispersal pattern). It is also possible that temporary
departures from HRs were not detected due to the
irregular nature of the tracking trips and the difficulty of relocating wide-ranging fish. In that case,
subsequent relocation of these same individuals at
previously occupied locations would indicate intraannual SF to established HRs. Therefore intra-annual
SF may be a key feature of adult female halibut
movement patterns in Glacier Bay during the summer and fall.
The study has also provided some evidence for
interannual SF for halibut in Glacier Bay. Of the 4 fish
released in November, 3 inhabited HRs at their
release locations the following summer. Whether or
not these fish left Glacier Bay during winter spawning migrations is unknown, but 2 of these fish were
observed at different locations within the park following tagging (thus demonstrating an SF dispersal
pattern).
These results complement previous observations of
SF for Pacific halibut from a pop-up satellite archival
tag (PSAT) study and provide further details on the
scales at which it may occur. Approx. 80% of summer-to-summer PSAT pop-up locations (n = 25) were
located within 20 km of release locations after 1 yr at
liberty (Loher 2008). Most (75%) of these fish had
returned to the release location following migrations
to deeper water in the Gulf of Alaska during winter,
presumably to spawn. Although the displacement from
the release locations from the PSAT study matches
the scale of the dispersive movement state observed
in this study, the demonstrated ability of fish in the
current study to return to within a few hundred
meters of their original locations after undertaking
forays indicates that SF for Pacific halibut likely
occurs at much finer spatial scales than can be
detected using PSATs. SF has also been observed for
many other flatfish species (Hunter et al. 2003, Solmundsson et al. 2005, Sackett et al. 2008, Dando
2011, Moser et al. 2013).
Random movement: diffusion. Although the majority of the fish in this study displayed non-random
movement patterns associated with an overall lack of
dispersal during summer, some fish did appear to
have more mobile movement patterns. The random
movement dispersal pattern demonstrated by a small
proportion of tagged halibut suggests that some halibut do not establish HRs, but may instead move randomly throughout summer foraging areas. The rate
of diffusion associated with random movement in this
study, 0.9 km2 d−1, is comparable to diffusion rates
estimated for other flatfish species such as Baltic Sea
turbot Psetta maxima (Florin & Franzén 2010) and
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
winter flounder Pseudopleuronectes americanus
(Saila 1961) based on results derived from conventional tag recaptures. Therefore, random movement
appears to be another common behavior of Pleuronectiformes species, likely as a foraging tactic.
However, sample sizes were low for this dispersal
pattern, so results should be interpreted with caution.
For example, these fish could also have been detected during temporary forays to or from HRs in
unknown locations.
A large-scale summer-to-summer PIT tag study of
67 000 halibut provided similar observations of both
sedentary and mobile movement patterns for adult
halibut that occurred over larger scales in space and
time. Fish tagged during 2003 and 2004 had not
mixed completely with the population by 2006 to
2009 (Webster et al. 2013) and as of 2008, 86% of
132 tags recaptured by annual survey vessels were
caught at the same survey station where they were
released (Loher 2008). Survey stations were located
on an 18.5 km grid, which matches the approximate
scale of the dispersive movement state observed in
Glacier Bay. These observations support the presence of a long-term sedentary movement pattern for
adult fish. On the other hand, the probability for
large-scale movement between management units
for large (e.g. 130 cm) fish was close to 20% for
some units (Webster et al. 2013), which suggests
that a more mobile movement pattern with a
greater potential for dispersal also exists for adult
halibut. In addition to the 4 fish that exhibited the
random dispersal pattern in our study, it is possible
that some of the 11 fish that were rarely or never
detected had more mobile movement patterns. In
that case, the proportion of tagged fish with more
mobile patterns would range from a minimum of
9% (4 of 43) to a maximum of 35% (15 of 43),
assuming no mortality, tag loss, or undetected HR
behavior at unknown locations within Glacier Bay
had occurred.
Caveats. Our use of a model selection framework to
link observed patterns of NSD vs. time to theoretical
models of dispersal represents a promising approach
for identifying and quantifying fish movement patterns in terms of ecological phenomena such as HR
occupation, foraging, and migration. This analysis
method is appropriate for data collected at irregular
intervals because the analysis is based on positive
observations of NSD at a given point in time. However, due to the small sample sizes obtained for fish
with more mobile movement patterns in this study,
the results for dispersal patterns other than HR
should be viewed as providing a preliminary under-
245
standing of the types of behavior and spatial scales
of movement that fish may demonstrate during
summer.
Habitat relationships
The habitat associations observed for tagged fish
may be related to a tendency for tagged fish to occupy a specific benthic habitat type in Glacier Bay.
Three regions composed of different combinations of
depth, tidal velocity, substrate type, and community
composition exist in Glacier Bay (Etherington et al.
2007a). The mouth and lower portions of Glacier Bay
consist of a large, flat, shallow (50 m), high-current
area with sand and cobble substrate associated with
a community of horse mussels, scallops, and sea
urchins. In contrast, the central and northern portions
of the bay are composed primarily of deep fjords (to
approx. 450 m) with muddy substrates (Fig. S4) and
were associated with Tanner crab (Chionoecetes
bairdi), shrimp, and flatfish species. However, the
majority of fish in this study were tagged and tracked
in a transition zone between these 2 areas that is
characterized by intermediate depths, intermediate
levels of tidal velocity, mixed cobble/soft sediment,
and intermediate to high levels of habitat complexity.
This region is also occupied by Pacific herring (Clupea palasii), Pacific cod (Gadus macrocephalus),
walleye pollock (Gadus chalcogrammus), rockfishes
(Sebastes spp.), and other common prey items for
halibut (Best & St-Pierre 1986, Etherington et al.
2007a, Moukhametov et al. 2008, Renner et al. 2012).
This transition area is also a highly productive front
where well-mixed water from the mouth of the bay
meets nutrient-rich stratified waters from the fjords
(Etherington et al. 2007b).
Significant associations between the residential
movement state and measures of habitat heterogeneity (change-in-slope, rugosity) and tidal velocity
may also be related to a sit-and-wait foraging strategy. For example, complex habitat can aid concealment during ambush and tides may deliver pelagic
prey (see Beaudreau & Essington 2011) to ambush
predators. The strongest habitat association observed was for the change-in-slope variable, which represents interfaces between shallow and steep
slopes as well as areas where depth is frequently
changing. Tagged halibut tended to be found in close
proximity to high values of the change-in-slope variable where glacial features such as moraines and ice
scours interface with large expanses of flat, homogeneous terrain in the lower portion of Glacier Bay
246
Mar Ecol Prog Ser 517: 229–250, 2014
(Fig. S5). Associations with interfaces between different habitat types have been observed for other fish
species such as the barred sand bass Paralabrax
nebulifer which inhabited interfaces between rocky
reefs used for hunting and adjacent soft-sediment
habitats used for resting or refuge (Mason & Lowe
2010). Although flatfishes are often associated with
soft sediments related to their tendency to bury in
sediments (Gibson 2005), no significant habitat association with distance to soft substrate habitats was
observed here, a result that could be related to the
abundance of soft sediment in the study area or a
reduced tendency to bury in sediment for adult fish
compared to juveniles.
Of the environmental and biological explanatory
variables examined, only depth was significantly related to HR size. Increased scales of movement in
deeper areas could reflect differences in prey type or
prey densities compared to shallow areas (i.e. transition region communities compared to deep fjord bottom communities, as discussed above). A positive
relationship with depth has also been observed for
temperate reef-associated fishes, where species that
occupy deeper depths tend to have larger ranges of
movement than those that occur at shallower depths
(Freiwald 2012). An increase in telemetry error with
depth could confound the relationship between
depth and HR size; however, the expected telemetry
error is still small (100 m) relative to the distances
moved in the larger HRs (1 to 2 km). Note that the
95% CIs reported for the GAM do not include errors
associated with uncertainty of position related to
depth. Because this model explained only 30% of the
variance, HR is probably affected by variables that
were either not measured or occurred at different
scales in space or time. For example, the movements
of many fish species are known to be related to tidal
patterns on a daily basis (Tolimieri et al. 2009), so the
use of time-and-depth averaged tidal velocity in this
study may have been too coarse to detect relationships with tide. Although positive relationships between fish length and HR size have been previously
reported (Kramer & Chapman 1999), we found that
HR size was not related to fish size in this study.
However the size distribution of the fish tagged in
this study was relatively homogeneous (Fig. S1), so
this result could also be related to low numbers of
very small or very large fish. Finally, HR size did not
change over time based on the lack of significance of
the year variable. This result implies a stability of HR
scales over time as well as a lack of effect of changing
tag types (frequency, size, longevity, attachment
method) as the study evolved (Table S2). Recent
fieldwork by the authors (J. K. Nielsen & A. C. Seitz
unpubl. data), where 15 adult female halibut were
tracked for 2 mo in Glacier Bay, has provided independent confirmation of the scale and dominance of
the residential movement pattern during summer.
Implications for MPAs and spatial fisheries
management
Determining fish movement scales relative to MPA
size is one of the most important aspects of MPA design (Gruss et al. 2011, Saarman et al. 2013). Scales of
both residential and dispersive movement states
were smaller than the scale of the Glacier Bay MPA,
and most tagged fish were detected regularly inside
the MPA boundary. Thus, Glacier Bay is likely to
encompass the majority of movements of individual
adult female Pacific halibut during the summer and
fall. Retention of halibut within Glacier Bay may be
encouraged by the enclosed nature of the bay, availability of heterogeneous habitat with which tagged
halibut were associated, and the productivity of its
waters. Glacier Bay would therefore be expected to
serve as a refuge from commercial harvest after the
phase-out of commercial fisheries in the park is completed (estimated by the National Park Service to
occur sometime between 2040 and 2050). However,
understanding the potential for specific benefits of
Glacier Bay as an MPA such as change in size structure or abundance (Taggart et al. 2004) will require
more information on (1) large-scale movement patterns of halibut over yearly timescales, as fish could
be vulnerable to commercial fishing during migrations from summer foraging locations to winter
spawning locations outside of Glacier Bay, and (2)
the effect of ongoing charter and unguided sport
fishing within the park that will continue after commercial fishing is phased out.
In addition to insights into the potential utility of
Glacier Bay as an MPA, this study has also yielded information that may be useful for design of MPAs or
spatial management in general for halibut. The frequent observations of non-random dispersal patterns
such as HR and SF are strikingly similar to dispersal
patterns observed for other temperate reef-associated
fishes such as lingcod (Ophiodon elongatus) and
some rockfish species (Matthews 1990, 1992, Pearcy
1992, Starr et al. 2004, Tolimieri et al. 2009, Beaudreau & Essington 2011). Lingcod, yellowtail rockfish
(Sebastes flavidus), blue rockfish (Sebastes mystinus),
and California scorpionfish (Scorpaena guttata) were
found to exhibit non-dispersive movement at spatial
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
scales that were similar to tagged halibut in this study
(Freiwald 2012, his Supplement 1, Fig. S1). Benefits
such as increases in biomass and egg production
have been observed for multiple temperate reef fish
species within MPAs in California (Tetreault & Ambrose 2007). Movement patterns and scales for adult
female lingcod tagged in a small MPA near Sitka,
Alaska, which were very similar to those of the adult
female halibut tagged in our study, suggest that
MPAs may facilitate increased egg production through
protection of lingcod brood stock (Starr et al. 2004).
Our research suggests that MPAs may also provide
some degree of protection for Pacific halibut brood
stock, and therefore potential enhancement of egg
production, based on observations of HR and SF during the summer and fall. The effectiveness of such
MPAs would then depend on the timing of winter
spawning migrations, which may occur just before
and/or after the commercial fishing season (Loher
2011), as well as the proportion of adult females that
undertake annual spawning migrations (Loher &
Seitz 2008). As the estimated total biomass of Pacific
halibut in the eastern North Pacific Ocean has declined by 50% between 1996 and 2013 and the majority of fish captured in the commercial fishery are females (Stewart et al. 2013), additional time-area
closures in areas where large females are found in
high abundance could provide some measure of protection during periods of declining stock abundance.
However, additional research will be required to
determine the extent to which the movement scales
and habitat associations observed in Glacier Bay, a
relatively enclosed fjord estuary, may be applicable
to other marine environments that occur throughout
the range of halibut distribution. For example, limited
movement within fjords has been observed for Atlantic halibut (Seitz et al. 2014) and Atlantic cod
(Hedger et al. 2011). Movement scales may also be
related to the availability of complex habitat and high
relief areas in a given location (Matthews 1990, Beaudreau & Essington 2011). Our results indicate that in
Glacier Bay, scales of movement tend to be smaller in
shallow (<100 m) areas with heterogeneous topography and complex habitat compared to deep, flat areas
with low levels of habitat complexity. Thus, although
some locations are likely to be more effective than
others at retaining the movements of halibut per unit
area, more research on halibut movement scales in
different habitat types and geographic regions is
needed to determine how much of the sedentary
movement patterns observed in this study are due to
(1) fjord topography, (2) the presence of complex
habitat, or (3) inherent behavior of adult halibut.
247
Finally, the existence of fine-scale SF for some proportion of reproductive females has implications for
depletion of mature female fish at local scales.
Although the phenomenon of local depletion for
Pacific halibut has not been explicitly documented,
some evidence from a variety of sources suggests
that it may be of concern. For example, declines in
commercial catch per unit effort near the Pribilof
Islands occurred in conjunction with concentrated
local fishing effort (Hare 2005). Localized declines
have also been observed near populated areas,
where intense charter and sport fishing effort occurs
(Trumble et al. 1991). Our findings of HR behavior
combined with SF and the ability to return to previously occupied areas suggest a potential mechanism
by which Pacific halibut could be vulnerable to local
depletion. However, before the potential for local
depletion can be characterized, broad geographicscale dispersal processes, population connectivity,
and spatial structure during other life history phases,
such as passive planktonic larval drift and contranatant
ontogenetic migrations, must be fully assessed (Skud
1977, Conners & Munro 2008).
Benefits of NSD analysis methods
Our use of NSD analyses to describe and quantify
movement patterns and scales is a novel, robust approach that can be applied to irregular datasets commonly collected for fishes. Using the analysis of dispersal patterns, telemetry records for tagged fish can
be analyzed in the context of movement ecology and
potential for dispersal, rather than focusing on the
size or location of an area inhabited by the fish. The
mixed-effects model framework allows the description of inherent individual variability, as well as borrowing ‘strength’ from individuals with more observations. Information on movement characteristics is
provided in formats such as diffusion rates or HR
scales that can be easily used to simulate movement
paths for use in other studies. For example, such simulations could be used to estimate energy budgets
during residential versus dispersive movement states.
In addition, it provides a way to compare results between acoustic and conventional tagging studies. For
example, diffusion coefficients have been calculated
for American lobsters using the slope of NSD vs. time
from conventional tag recovery data (den Heyer et al.
2009). Thus, this method may provide a way to leverage the detailed information provided by acoustic
studies with the larger sample sizes available from
conventional tagging studies.
Mar Ecol Prog Ser 517: 229–250, 2014
248
Carlson PR, Hooge P, Cochrane G, Stevenson A, Dartnell P,
Finally, the NSD analysis framework complements
Lee K (2002) Multibeam bathymetry and selected perinformation needs for MPA design, where managers
spective views of main part of Glacier Bay, Alaska.
are often faced with the task of compiling informaReport no. 02-391. US Geological Survey, Menlo Park,
tion on the movement scales of multiple fish species
CA
Clark WG, Hare SR (2006) Assessment and management of
based on data collected using different methods (SaarPacific halibut: data, methods, and policy. Sci Rep IPHC
man et al. 2013). Because HR scales are reported as a
83
movement radius rather than area, NSD methods can
Conners ME, Munro P (2008) Effects of commercial fishing
be used to compare scales for fish that move in one
on local abundance of Pacific cod (Gadus macrodimension (e.g. in a river or along a coastline) with
cephalus) in the Bering Sea. Fish Bull 106:281−292
Conover WJ (1999) Practical nonparametric statistics. John
those that move in 2 dimensions. Combining differWiley & Sons, New York, NY
ent types of movement data (e.g. large-scale acoustic
Dando PR (2011) Site fidelity, homing and spawning migra➤
arrays, archival tags, or conventional tags) would also
tions of flounder Platichthys flesus in the Tamar estuary,
be possible because the only information required for
South West England. Mar Ecol Prog Ser 430:183−196
this method is distance moved from the release loca- ➤ den Heyer CE, Chadwick EMP, Hutchings JA (2009) Diffusion of American lobster (Homarus americanus) in
tion over time. Therefore this analysis approach may
Northumberland Strait, Canada. Can J Fish Aquat Sci
be particularly valuable because it provides move66:659−671
ment data in formats that can be easily combined or
Espasandin C (2012) Movement behavior of the California
halibut in a restored southern California wetland. MSc
compared with results from other studies.
Acknowledgements. We thank the National Park Service,
the US Geological Survey, the Rasmuson Fisheries Research
Center, the Pollock Conservation Cooperative Research
Center, and the North Pacific Research Board for financial
support of this project. Many people assisted with the fieldwork and data processing for this project, including Jim de
La Bruere, Elizabeth Hooge, Gretchen Bishop, Liz Chilton,
Caroline Dezan, Fritz Koschmann, Liz Solomon, and others.
We thank Milo Adkison, Franz Mueter, Tim Loher, Susanne
McDermott, Anne Beaudreau, Chad Soiseth, and Craig
Murdoch for providing valuable advice and guidance on the
manuscript.
➤
➤
LITERATURE CITED
➤ Austin D, Bowen WD, McMillan JI (2004) Intraspecific vari➤
➤
➤
➤
ation in movement patterns: modeling individual behaviour in a large marine predator. Oikos 105:15−30
Bartumeus F, Da Luz MGE, Viswanathan GM, Catalan J
(2005) Fish search strategies: a quantitative randomwalk analysis. Ecology 86:3078−3087
Beaudreau AH, Essington TE (2011) Use of pelagic prey
subsidies by demersal predators in rocky reefs: insight
from movement patterns of lingcod. Mar Biol 158:471−483
Best EA, St-Pierre G (1986) Pacific halibut as predator and
prey. Tech Rep IPHC 21:1−27
Blackwell PG (1997) Random diffusion models for fish movement. Ecol Model 100:87−102
Börger L, Fryxell J (2012) Quantifying individual differences
in dispersal using net squared displacement. In: Clobert
J, Baguette M, Benton T, Bullock J (eds) Dispersal and
spatial evolutionary ecology. Oxford University Press,
Oxford, p 222−230
Bunnefeld N, Börger L, van Moorter B, Rolandsen CM,
Dettki H, Solberg EJ, Ericsson G (2011) A model-driven
approach to quantify migration patterns: individual,
regional and yearly differences. J Anim Ecol 80:466−476
Burnham KP, Anderson DR (1990) Model selection and
multimodal inference: an information-theoretic approach.
Springer Verlag, New York, NY
➤
➤
thesis, California State University, Long Beach, CA
ESRI (2011) ArcGIS Desktop, Release 10.1.
Etherington LL, Cochrane GR, Harney J, Taggart SJ and
others (2007a) Glacier Bay seafloor habitat mapping and
classification — first look at linkages with biological patterns. In: Piatt JF, Gende SM (eds) 4th Glacier Bay Sci
Symp. US Geological Survey, Juneau, AK, p 71−75
Etherington LL, Hooge PN, Hooge ER, Hill DF (2007b)
Oceanography of Glacier Bay, Alaska: implications for
biological patterns in a glacial fjord estuary. Estuar Coast
30:927−944
Florin AB, Franzén F (2010) Spawning site fidelity in Baltic
Sea turbot (Psetta maxima). Fish Res 102:207−213
Freiwald J (2012) Movement of adult temperate reef fishes
off the west coast of North America. Can J Fish Aquat Sci
69:1362−1374
Geiselman J, Dunlap J, Hooge P, Albert D (1997) Glacier
Bay ecosystem GIS CD-ROM set. US Geological Survey
and Interrain Pacific, Anchorage and Juneau, AK
Gibson RN (2005) The behaviour of flatfishes. In: Gibson RN
(ed) Flatfishes: biology and exploitation. Blackwell Science, Oxford, p 213−239
Giuggioli L, Bartumeus F (2012) Linking fish movement to
site fidelity. J Math Biol 64:647–656
Gruss A, Kaplan DM, Guenette S, Roberts CM, Botsford LW
(2011) Consequences of adult and juvenile movement for
marine protected areas. Biol Conserv 144:692−702
Haaker P (1975) The biology of the California halibut, Paralichthys californicus (Ayres) in Anaheim Bay. In: Lane
ED, Hill CW (eds) The marine resources of Anaheim Bay,
Vol 165. Calif Dep Fish Game Fish Bull, p 137−151
Hare SR (2005) Investigation of the role of fishing in the
Area 4C CPUE decline. In: Sadorus L (ed) Report of
assessment and research activities 2004. IPHC, Seattle,
WA, p 185−197
Harney JN, Cochrane GR, Etherington LL, Dartnell P,
Golden NE, Chezar H (2006) Geologic characteristics of
benthic habitats in Glacier Bay, southeast Alaska. Report
no. 2006-1081, US Geological Survey, Santa Cruz, CA
Hedger RD, Uglem I, Thorstad EB, Finstad B and others
(2011) Behaviour of Atlantic cod, a marine fish predator,
during Atlantic salmon post-smolt migration. ICES J Mar
Sci 68:2152−2162
Nielsen et al.: Movement of Pacific halibut in a Marine Protected Area
249
➤ Hill DF, Ciavola SJ, Etherington L, Klaar MJ (2009) Estima- ➤ Nilsson J, Kristiansen TS, Fosseidengen JE, Stien LH, Ferno
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
tion of freshwater runoff into Glacier Bay, Alaska and
incorporation into a tidal circulation model. Estuar Coast
Shelf Sci 82:95−107
Hunter E, Metcalfe JD, Reynolds JD (2003) Migration route
and spawning area fidelity by North Sea plaice. Proc R
Soc Lond B Biol Sci 270:2097−2103
Jorgensen SJ, Reeb CA, Chapple TK, Anderson S and others
(2010) Philopatry and migration of Pacific white sharks.
Proc R Soc Lond B Biol Sci 277:679−688
Kareiva PM, Shigesada N (1983) Analyzing insect movement
as a correlated random walk. Oecologia 56:234−238
Kramer DL, Chapman MR (1999) Implications of fish home
range size and relocation for marine reserve function.
Environ Biol Fishes 55:65−79
Loher T (2008) Homing and summer feeding site fidelity of
Pacific halibut (Hippoglossus stenolepis) in the Gulf of
Alaska, established using satellite-transmitting archival
tags. Fish Res 92:63−69
Loher T (2011) Analysis of match−mismatch between commercial fishing periods and spawning ecology of Pacific
halibut (Hippoglossus stenolepis), based on winter surveys and behavioural data from electronic archival tags.
ICES J Mar Sci 68:2240−2251
Loher T, Rensmeyer R (2011) Physiological responses of
Pacific halibut, Hippoglossus stenolepis, to intracoelomic
implantation of electronic archival tags, with a review of
tag implantation techniques employed in flatfishes. Rev
Fish Biol Fish 21:97−115
Loher T, Seitz A (2006) Seasonal migration and environmental conditions of Pacific halibut Hippoglossus stenolepis,
elucidated from pop-up archival transmitting (PAT) tags.
Mar Ecol Prog Ser 317:259−271
Loher T, Seitz AC (2008) Characterization of active spawning season and depth for eastern Pacific halibut (Hippoglossus stenolepis), and evidence of probable skipped
spawning. J Northwest Atl Fish Sci 41:23−36
Mason TJ, Lowe CG (2010) Home range, habitat use, and
site fidelity of barred sand bass within a southern California marine protected area. Fish Res 106:93−101
Matthews KR (1990) A telemetric study of the home ranges
and homing routes of copper and quillback rockfishes on
shallow rocky reefs. Can J Zool 68:2243−2250
Matthews KR (1992) A telemetric study of the home ranges
and homing routes of lingcod Ophiodon elongatus on
shallow rocky reefs off Vancouver Island, British Columbia. Fish Bull 90:784−790
Moorcroft PR, Lewis MA (2006) Mechanistic home range
analysis. Princeton University Press, Princeton and
Oxford
Morales JM, Haydon DT, Frair J, Holsinger KE, Fryxell JM
(2004) Extracting more out of relocation data: building
movement models as mixtures of random walks. Ecology
85:2436−2445
Moser ML, Myers MS, West JE, O’Neill SM, Burke BJ (2013)
English sole spawning migration and evidence for feeding site fidelity in Puget Sound, U.S.A., with implications
for contaminant exposure. Northwest Sci 87:317−325
Moukhametov IN, Orlov AM, Leaman BM (2008) Diet of
Pacific halibut (Hippoglossus stenolepis) in the northwestern Pacific Ocean. Tech Rep IPHC 52
Nakano S, Fausch KD, Kitano S (1999) Flexible niche partitioning via a foraging mode shift: a proposed mechanism
for coexistence in stream-dwelling charrs. J Anim Ecol
68:1079−1092
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
➤
A, van den Bos R (2010) Learning and anticipatory
behaviour in a ‘sit-and-wait’ predator: the Atlantic
halibut. Behav Processes 83:257−266
Nouvellet P, Bacon JP, Waxman D (2009) Fundamental
insights into the random movement of fish from a single
distance-related statistic. Am Nat 174:506−514
Papworth SK, Bunnefeld N, Slocombe K, Milner-Gulland EJ
(2012) Movement ecology of human resource users:
using net squared displacement, biased random bridges
and resource utilization functions to quantify hunter and
gatherer behaviour. Methods Ecol Evol 3:584−594
Pearcy WG (1992) Movements of acoustically-tagged yellowtail rockfish Sebastes flavidus on Heceta bank, Oregon. Fish Bull 90:726−735
Perry RI, Smith SJ (1994) Identifying habitat associations of
marine fishes using survey data: an application to the
northwest Atlantic. Can J Fish Aquat Sci 51:589−602
Pinheiro J, Bates D (2000) Mixed-effects models in S and SPLUS. Springer-Verlag, New York, NY
Renner M, Arimitsu ML, Piatt JF (2012) Structure of marine
predator and prey communities along environmental
gradients in a glaciated fjord. Can J Fish Aquat Sci 69:
2029−2045
Rogers KB, White GC (2007) Analysis of movement and
habitat use from telemetry data. In: Brown M, Guy C
(eds) Analysis and interpretation of freshwater fisheries data. American Fisheries Society, Bethesda, MD,
p 625−676
Saarman E, Gleason M, Ugoretz J, Airamé S and others
(2013) The role of science in supporting marine protected
area network planning and design in California. Ocean
Coast Manag 74:45−56
Sackett DK, Able KW, Grothues TM (2008) Habitat dynamics of summer flounder Paralichthys dentatus within a
shallow USA estuary, based on multiple approaches
using acoustic telemetry. Mar Ecol Prog Ser 364:199−212
Saila SB (1961) A study of winter flounder movements. Limnol Oceanogr 6:292−298
Seitz AC, Loher T, Norcross BL, Nielsen JL (2011) Dispersal
and behavior of Pacific halibut Hippoglossus stenolepis
in the Bering Sea and Aleutian Islands region. Aquat Biol
12:225−239
Seitz AC, Michalsen K, Nielsen JL, Evans MD (2014) Evidence of fjord spawning by southern Norwegian Atlantic
halibut (Hippoglossus hippoglossus). ICES J Mar Sci 71:
142–147
Singh NJ, Borger L, Dettki H, Bunnefeld N, Ericsson G
(2012) From migration to nomadism: movement variability in a northern ungulate across its latitudinal range.
Ecol Appl 22:2007−2020
Skud BE (1977) Drift, migration, and intermingling of Pacific
halibut stocks. Sci Rep IPHC 63
Solmundsson J, Palsson J, Karlsson H (2005) Fidelity of
mature Icelandic plaice (Pleuronectes platessa) to spawning and feeding grounds. ICES J Mar Sci 62:189−200
Starr RM, O’Connell V, Ralston S (2004) Movements of lingcod (Ophiodon elongatus) in southeast Alaska: potential
for increased conservation and yield from marine reserves. Can J Fish Aquat Sci 61:1083−1094
Staudinger MD, Juanes F (2010) Feeding tactics of a behaviorally plastic predator, summer flounder (Paralichthys
dentatus). J Sea Res 64:68−75
Stewart IJ, Leaman BM, Martell S, Webster RA (2013)
Assessment of the Pacific halibut stock at the end of
250
➤
➤
Mar Ecol Prog Ser 517: 229–250, 2014
2012. In: Sadorus L (ed) Report of assessment and
research activities 2012. IPHC, Seattle, WA, p 93−186
Taggart SJ, Shirley TC, O’Clair CE, Mondragon J (2004)
Dramatic increase in the relative abundance of large
male dungeness crabs Cancer magister following closure
of commercial fishing in Glacier Bay, Alaska. Am Fish
Soc Symp 42:243−253
Tetreault I, Ambrose RF (2007) Temperate marine reserves
enhance targeted but not untargeted fishes in multiple
no-take MPAs. Ecol Appl 17:2251−2267
Tolimieri N, Andrews K, Williams G, Katz S, Levin PS (2009)
Home range size and patterns of space use by lingcod,
copper rockfish and quillback rockfish in relation to diel
and tidal cycles. Mar Ecol Prog Ser 380:229−243
Trumble RJ, St-Pierre G, McGregor IR (1991) Evaluation of
Pacific halibut management for regulatory area 2A. Sci
Rep IPHC 74
Turchin P (1998) Quantitative analysis of movement: measuring and modeling population redistribution in fish and
plants. Sinauer Associates, Sunderland, MA
Editorial responsibility: Nicholas Tolimieri,
Seattle, Washington, USA
➤
Valero JL, Webster RA (2012) Current understanding of
Pacific halibut migration patterns. In: Sadorus L (ed)
Report of assessment and research activities 2011. IPHC,
Seattle, WA, p 341−379
Webster RA, Stewart IJ (2014) Apportionment and regulatory area harvest calculations. In: Sadorus L (ed) Report
of assessment and research activities 2013. IPHC, Seattle, WA, p 197−216
Webster RA, Clark WG, Leaman BM, Forsberg JE (2013)
Pacific halibut on the move: a renewed understanding of
adult migration from a coastwide tagging study. Can J
Fish Aquat Sci 70:642−653
Wood SN (2006) Generalized Additive Models: an introduction with R. Chapman & Hall/CRC, Boca Raton, FL
Wright DJ, Pendleton M, Boulware J, Walbridge S, Gerlt B,
Eslinger D, Sampson D, Huntley E (2012) ArcGIS Benthic
Terrain Modeler (BTM), v. 3.0. Environmental Systems
Research Institute, NOAA Coastal Services Center, Massachusetts Office of Coastal Zone Management. http://
esriurl.com/5754
Submitted: March 6, 2014; Accepted: September 12, 2014
Proofs received from author(s): December 3, 2014